NO. 4 | JULY 20, 2012
The Difficulty of Trading “Ultra-Liquid” Stocks
T
This schematic shows a limit order’s journey through the time priority queue of a limit order book, and how longer queues can interfere with passive trading.
time ASK BID
CROSS THE SPREAD TO CATCH UP
QUEUE
he shortfall of an order is affected by several wellknown characteristics—the order’s participation rate, the stock’s volatility and bid-ask spread. But another important factor is how often it is necessary to cross the bid-ask spread. In order to minimize execution costs, many traders use low-participation-rate algorithms allowing them to avoid crossing the spread often, thereby reducing the spread costs associated with taking liquidity. In general, the lower the participation rate of an order, the easier it is to trade, so it is intuitive that securities with extremely high average daily volumes are particularly easy to trade passively due to the ample supply of aggressive orders to execute against one’s passive orders. However, this is not always true. On the contrary, it is often more difficult to trade ultrahigh-volume names relative to lower-ADV stocks, and in particular, to provide liquidity. Providing less implies getting a worse price, by crossing the spread or by seeking a midpoint fill in lit or dark venues, which leads to higher shortfalls. In the following study, we demonstrate why the common belief about the ease of trading extremely high-ADV names such as those listed in the Appendix does not match the reality.
Volume ≠ Liquidity
P RAG M A R E S E A R C H NOTE
ASK FILLED AT THE BID
BID QUEUE
How can higher volume not lead to lower trading costs? Consider buying 1% of the ADV of CSCO using a TWAP strategy over a two-hour time frame. CSCO trades 47 million shares per day, which means a TWAP algo must execute about 4,000 shares each minute. To stay reasonably close to its target, the trading algorithm might be configured to cross the spread whenever it falls behind the target quantity by more than 2,000 shares, i.e. the unfilled quantity that results from lagging behind the target quantity by more than half a minute. Suppose, in the hope of not having to cross the spread, the algo posts some fraction of the order quantity on the bid at the end of a long queue of orders already waiting to get executed (84,000 shares based on the average value from the Appendix). During the first 15 seconds
MARKET ORDER FROM OPPOSING SIDE POSTED ORDER AHEAD OF ALGO ORDER POSTED ALGO ORDER ORDERS POSTED BEHIND ALGO ORDER
pragmat rading.com
1
12
VWA P S F (BPS )
10 8 6 4 2 0 10
32
FIGURE 1 Average VWAP shortfall versus the average time it takes to deplete the queue*.
100
320
1000
3200
Q UOT E D E P L E T I ON T I ME ( S)
QU OTE DE P LETIO N T I ME ( M I N )
of sitting on the queue, 15,000 shares have been Note that the details of this stylized example are taken from the book. The algo’s posted order has not essential—regardless of whether a trader uses not yet moved to the front of the queue and remains an algorithm or trades manually, qualitatively all 6.3causing the algo to fall behind its target unfilled, directional traders have a limited patience to provide by 1,000 shares. After another 15 seconds, though liquidity before crossing the spread. Other things 30,000 shares of CSCO have traded, the posted order being equal, a longer queue makes it more likely is still4.0 not executed, and continues to move along the that patience will be exhausted and the spread will queue. The algo is now 2,000 shares behind and is be crossed. Also note that there are good reasons forced to cross the spread to catch up. This pattern to have limited patience—routinely waiting too long may be repeated many times over the course of the before crossing the spread can lead to an adverse 2.5 trade, leading to a poor execution. selection effect, where stocks are allowed to run far In contrast, consider a similar scenario of trading 1% away before a trade is completed. The effect disof the ADV of a less liquid stock, EXPD, (ADV of 2.7M cussed in this paper is not just an artifact of unreasonshares), again using a two-hour TWAP algo, at a rate ably impatient traders but rather is characteristic of 1.6 of 225 shares per minute. EXPD has an average queue sensible trading approaches. length of only 1,100 shares. After 20 seconds, when Figure 1 illustrates the effect at a statistical level, the algo’s order has worked through the queue and using performance data from actual orders. As preis about0to be executed, the algo is behind by only a dicted, the average VWAP shortfall is an increasing fraction of a lot (75 shares). The posted shares, in this function of the time it takes to deplete the queue. 0.1 without having 0.3 to cross the 1.0spread, Queue 3.2 depletion10.0 31.6to the average 100.0 case, are executed time is proportional leading to better shortfall. The queue length, therequote size for the stock divided by the average daily AV ERAGE D AI LYvolume VOL UME ( M SH AR E S) fore, has a strong effect on performance. of the stock.
* The blue dotted lines represent a 95% confidence interval
6
LL ( BPS )
JU LY 2 0 , 2 0 1 2
pragmat rading.com
2
10
32
320 Q U O100 TE DEPLETIO N T I M E ( S ) 1000
3200
A longer queue depletion time leads to a lower provide rate, and thereby to paying the full or half the spread more often, which results in worse execution shortfall. At first this seems consistent with the common intuition that higher volume means cheaper trading, but in fact this intuition is accurate only if queue length is held equal. Figure 2 shows the queue depletion time does not simply drop as volume increases. The queue depletion time does initially fall as volume grows, but at a certain point—around 5 million shares/ day—average queue length starts growing even faster than volume, as shown by the higher queue depletion time. Correspondingly, Figure 3 shows an initial decline in the shortfall of individual child orders from first placement to eventual execution. This is at least consistent with the common intuition that higher volume stocks are easier to trade than low-volume names. However at still higher volumes, as the queue depletion times grow, it becomes increasingly hard to provide liquidity while staying within a discretionary range of some target (set by POV, TWAP, VWAP, etc.) and shortfalls increase accordingly. While not shown here, there is also a clear relation between the time to deplete the queue and the percentage of the orders executed via aggressive orders. Putting together all these results shows that, counter-intuitively, on average highervolume stocks have disproportionately long queues and are costlier to trade.
6.3 6.3 4.0 4.0 2.5 2.5 1.6 1.6 0 0
0.1
0.3
0.1
0.3
1.0
3.2
10.0
1.0 AV E R A G E D A I LY V3.2 O L U M E ( M10.0 SHARES)
31.6
100.0
31.6
100.0
AV E R A G E D A I LY V O L U M E ( M S H A R E S )
F IGU RE 2 Average queue depletion time versus ADV*. C H I LCDH O I LR DDO E R DSEHROSRT HO FART L LFA ( BP L L S()B P S )
Volume, Queue Length and Shortfall
QU OT QU EOT D EEP D L EETPILON E T IT ON I MET I(ME MI N( MI ) N)
QUOTE DEPLETION TIME (S)
6 6 5 5 4 4 3 3 0.1
0.3
0.1
0.3
1.0
3.2
10.0
1.0 AV E R A G E D A I LY 3.2 V O L U M E ( M10.0 SHARES)
31.6
100.0
31.6
100.0
AV E R A G E D A I LY V O L U M E ( M S H A R E S )
F IGU RE 3 Average child order shortfall versus ADV*.
Conclusion Although volume and liquidity are often thought of as synonymous, the large number of shares typically posted on the book of very high volume stocks makes it difficult to gain priority and get executed with limit orders. In the race to stay close to a benchmark and avoid adverse selection, traders must get more aggressive to escape the crowded book of competing quotes, and suffer the consequences in worse shortfall.
For questions or comments, please email Dr. Eran Fishler, Chief Operating Officer (technotes@pragmatrading.com). Past performance is not a guarantee of future returns. Copyright © 2012 Pragma Securities. All rights reserved. Do not reproduce or excerpt without permission. Pragma Securities, LLC. Member of FINRA, NASDAQ and SIPC. C.A. #83
JU LY 2 0 , 2 0 1 2
pragmat rading.com
3
TICKER
Appendix
50 Largest ADV Stocks as of June 2012
JU LY 2 0 , 2 0 1 2
ADV (Mshares)
Queue Length (Shares)
Spread (cents)
15-min Volatility (%)
Queue Depletion Time (s)
BAC
190.3
558,925
1.0
0.41
SPY
161.1
30,237
1.0
0.12
9
XL F
89.6
2,574,847
1.0
0.19
1345
138
FB
74.5
22,296
1.3
0.91
14
EEM
60.9
82,727
1.0
0.16
64
JPM
56.8
13,250
1.0
0.30
11
VXX
56.4
11,852
1.0
0.57
10 920
SIRI
53.8
1,021,571
1.0
0.45
QQQ
50.2
63,295
1.0
0.15
59
IWM
48.7
15,158
1.0
0.19
15
C SC O
47.3
84,527
1.0
0.23
85
MSFT
45.6
40,857
1.0
0.20
42
F
43.8
132,213
1.0
0.31
141
GE
43.8
89,467
1.0
0.20
96
S
43.5
422,473
1.0
0.58
456
C
41.9
14,347
1.0
0.37
16
C HK
38.6
10,778
1.0
0.52
13
INTC
36.9
35,485
1.0
0.22
45
NO K
32.2
430,861
1.0
0.37
627
PFE
30.9
58,008
1.0
0.19
88
MS
29.7
24,323
1.0
0.43
38
WFC
29.4
14,728
1.0
0.26
23
ORC L
29.3
22,414
1.0
0.26
36 159
MU
27.9
94,524
1.0
0.52
Z NGA
25.7
16,315
1.0
0.82
30
T
25.4
33,937
1.0
0.14
63
DELL
23.6
53,863
1.0
0.26
107
VWO
22.8
43,736
1.0
0.15
90
VAL E
21.9
23,396
1.0
0.31
50
TZA
21.8
11,200
1.1
0.56
24
EFA
21.8
34,171
1.0
0.14
73
E MC
21.7
23,829
1.0
0.25
52
SD S
21.0
316,091
1.0
0.24
706
AA
21.0
100,122
1.0
0.32
224
HPQ
20.7
20,302
1.0
0.27
46
GD X
20.3
4,984
1.1
0.32
11
E WZ
20.3
5,748
1.0
0.22
13
RF
20.2
134,203
1.0
0.45
311
L OW
20.0
11,869
1.0
0.26
28
FA Z
19.3
7,236
1.1
0.50
18
FXI
19.2
22,049
1.0
0.17
54
YHOO
18.9
36,246
1.0
0.29
90
XL I
17.7
169,464
1.0
0.16
447 71
NWSA
17.5
26,692
1.0
0.24
AAPL
17.5
416
14.3
0.25
1
ARNA
17.0
24,373
1.0
0.98
68
XL E
16.6
8,623
1.0
0.19
24
TVIX
16.2
5,162
1.2
1.04
15
PBR
16.2
16,496
1.0
0.33
48
RIMM
16.0
19,069
1.0
0.53
56
pragmatrading.com
4